{
  "id": 82096,
  "title": "-33 users? ",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/82096",
  "author_name": "",
  "post_date": "2019-02-27T07:05:03.348811Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hey. </p>\n\n<p>Our team spent a lot of time figuring out what was the nature of -33 users. </p>\n\n<p>At first, we thought about churn, as it was written in the forum. We defined churn as an absence of transactions for a given user for the next 2 months. We tried to get more churns by augmenting train dataset (did not improve LB score), and we noticed that the number of churn users was about 10% compared to the fraction of -33 users - 1%. </p>\n\n<p>About 2 weeks before the end of the competition, I came up with the idea that those users might be some kind of fraud. However, we did not find any evidence to support this theory. </p>\n\n<p>What do you think? </p>",
  "messages": [
    {
      "id": "479471",
      "postDate": "02/27/2019 07:05:03",
      "content": "<p>Hey. </p>\n\n<p>Our team spent a lot of time figuring out what was the nature of -33 users. </p>\n\n<p>At first, we thought about churn, as it was written in the forum. We defined churn as an absence of transactions for a given user for the next 2 months. We tried to get more churns by augmenting train dataset (did not improve LB score), and we noticed that the number of churn users was about 10% compared to the fraction of -33 users - 1%. </p>\n\n<p>About 2 weeks before the end of the competition, I came up with the idea that those users might be some kind of fraud. However, we did not find any evidence to support this theory. </p>\n\n<p>What do you think? </p>",
      "rawMarkdown": "Hey. \n\nOur team spent a lot of time figuring out what was the nature of -33 users. \n\nAt first, we thought about churn, as it was written in the forum. We defined churn as an absence of transactions for a given user for the next 2 months. We tried to get more churns by augmenting train dataset (did not improve LB score), and we noticed that the number of churn users was about 10% compared to the fraction of -33 users - 1%. \n\nAbout 2 weeks before the end of the competition, I came up with the idea that those users might be some kind of fraud. However, we did not find any evidence to support this theory. \n\nWhat do you think?",
      "votes": null
    },
    {
      "id": "479474",
      "postDate": "02/27/2019 07:11:00",
      "content": "<p>take a look at this:</p>\n\n<p><a href=\"https://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target\">https://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target</a></p>\n\n<p>it supports your churn theory</p>",
      "rawMarkdown": "take a look at this:\n\nhttps://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target\n\n\nit supports your churn theory",
      "votes": null
    },
    {
      "id": "479477",
      "postDate": "02/27/2019 07:15:52",
      "content": "<p>My response might be irrelevant to your discussion, I have not spent enough time on this competition.  Firstly congratulations <a href=\"/tearth\">@tearth</a> . I feel you are a true master in doing F.E. You getting top in both Analytics Vidhya and Kaggle inspired me a lot.</p>",
      "rawMarkdown": "My response might be irrelevant to your discussion, I have not spent enough time on this competition.  Firstly congratulations @tearth . I feel you are a true master in doing F.E. You getting top in both Analytics Vidhya and Kaggle inspired me a lot.",
      "votes": null
    },
    {
      "id": "479491",
      "postDate": "02/27/2019 07:35:03",
      "content": "<p>Thank you <a href=\"/karthik7395\">@karthik7395</a> :)</p>",
      "rawMarkdown": "Thank you @karthik7395 :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 479474,
      "author_name": "raddar",
      "author_url": "",
      "post_date": "02/27/2019 07:11:00",
      "content": "<p>take a look at this:</p>\n\n<p><a href=\"https://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target\">https://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target</a></p>\n\n<p>it supports your churn theory</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 479477,
      "author_name": "karthik7395",
      "author_url": "",
      "post_date": "02/27/2019 07:15:52",
      "content": "<p>My response might be irrelevant to your discussion, I have not spent enough time on this competition.  Firstly congratulations <a href=\"/tearth\">@tearth</a> . I feel you are a true master in doing F.E. You getting top in both Analytics Vidhya and Kaggle inspired me a lot.</p>",
      "votes": null,
      "replies": [
        {
          "id": 479491,
          "author_name": "",
          "author_url": "",
          "post_date": "02/27/2019 07:35:03",
          "content": "<p>Thank you <a href=\"/karthik7395\">@karthik7395</a> :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "479471": "Hey. \n\nOur team spent a lot of time figuring out what was the nature of -33 users. \n\nAt first, we thought about churn, as it was written in the forum. We defined churn as an absence of transactions for a given user for the next 2 months. We tried to get more churns by augmenting train dataset (did not improve LB score), and we noticed that the number of churn users was about 10% compared to the fraction of -33 users - 1%. \n\nAbout 2 weeks before the end of the competition, I came up with the idea that those users might be some kind of fraud. However, we did not find any evidence to support this theory. \n\nWhat do you think?",
    "479474": "take a look at this:\n\nhttps://www.kaggle.com/raddar/unobserved-data-simulation-exploiting-target\n\n\nit supports your churn theory",
    "479477": "My response might be irrelevant to your discussion, I have not spent enough time on this competition.  Firstly congratulations @tearth . I feel you are a true master in doing F.E. You getting top in both Analytics Vidhya and Kaggle inspired me a lot.",
    "479491": "Thank you @karthik7395 :)"
  },
  "source": "meta"
}